CN101697278A - Method for measuring music emotion distance - Google Patents

Abstract

The invention discloses a method for measuring music emotion similarity, which comprises the following steps: receiving a music segment; analyzing the emotional elements of the music segment; determining the main emotional element of the music segment; repeating the previous steps to process all pieces of music likewise; classifying music segments; calculating the emotion distance between music segments; and determining music emotion similarity according to the emotion distance between the music segments. By calculating the emotion distance between pieces of music, the similarity between the pieces of music can be measured accurately, and thus an effective means is provided for searching pieces of music having similar emotions.

Description

The measure of music emotion distance

[technical field]

The present invention relates to technical field of information processing, relate in particular to a kind of measure of weighing the music emotion distance of the emotion similarity between the music.

[background technology]

Music has become one of indispensable entertainment information in people's daily life, occupies an important position.The quantity of music is also increasing gradually, and number is more and more huger.Different users often likes different genres of music, has promptly comprised the music of different emotions.And sometimes when having different moods, also need to listen some similar music of mood at that time that suit.How to find suitable music in thousands of music, be the problem of an outbalance.

Music searching all is provided in bigger search website at present, provide a kind of " choosing song " service as the google music, the attribute of music is described from three aspects: rhythm is from releiving strongly, tone is from overcast to loud and sonorous, tone color is from enriching to gradually changing merely, and making up the music that has certain emotion to choose, this retrieval mode often has certain randomness.And for example Baidu's music provides the retrieval service that is called " music mood ", and music is classified according to sentimental, lonely, happy, warm and passion etc.

Above-mentioned music assorting or search method all have bigger subjectivity, can not accurately describe the emotion similarity between the music.

[summary of the invention]

Given this, be necessary to have bigger subjectivity at existing music emotion classification and selection method, can not comparatively accurately describe the problem of the emotion similarity between the music, a kind of balancing method of weighing the music emotion similarity of the emotion similarity between the music is provided.

A kind of balancing method of happy emotion similarity comprises the steps: to receive snatch of music; Analyze the emotion composition of snatch of music: determine the emotion disaggregated model, and analyze the emotion composition of described snatch of music according to described emotion disaggregated model, determine the shared number percent of each emotion composition in the described snatch of music, obtain with described number percent be element, the music emotion vector that is associated with described snatch of music; Determine the main emotion composition of snatch of music: with the emotion composition that accounts for the largest percentage main emotion composition as described snatch of music; Repeat above-mentioned steps all music are carried out same processing; Snatch of music is classified: the identical snatch of music of main emotion composition is divided into a class; Emotion distance between the computational music fragment: to belonging to the distance between the music emotion vector that of a sort snatch of music calculates with each snatch of music is associated; Determine the music emotion similarity according to the emotion distance between snatch of music: emotion distance is big more, and similarity is more little; Emotion distance is more little, and similarity is big more.

By the emotion distance between the computational music, can accurately weigh the emotion similarity between the music, for the music retrieval with similar emotion provides a kind of effective means.

Preferably, described emotion disaggregated model is Hevner emotion ring model or thayer thymopsyche model.

Preferably, the emotion distance between the computational music fragment comprises the Euclidean distance that calculates between the music emotion vector be associated with snatch of music, promptly use D (i, j) _IRepresent i snatch of music X in the I class emotion _iWith j snatch of music X _jDistance:

D(i，j) _I＝dis(P _i，P _j)

X wherein _i, X _jThe music of ∈ I class emotion, dis is the function of computed range, P _i, P _jBe respectively X _i, X _jThe emotion composition, s is the emotion classification number in the emotion disaggregated model that adopts:

P _i＝[a _i，1，a _i，2，...，a _i，s，]

P _j＝[a _j，1，a _j，2，...，a _j，s，]

Then the Euclidean distance between the music emotion vector calculates with following formula:

$\mathrm{dis}(P_i,P_j)=\sqrt{\underset{m=1}{\overset{s}{\mathrm{\Σ}}}{(a_{i,m}-b_{j,m})}^2}.$

[description of drawings]

Fig. 1 is the process flow diagram of computational music emotion distance;

Fig. 2 is a Hevner emotion ring synoptic diagram.

[embodiment]

As shown in Figure 1, be the process flow diagram of computational music emotion distance.Specifically comprise step:

S10: receive snatch of music.Snatch of music is meant one section music that the emotion composition is more stable.For several minutes song of a head, its emotion keynote is generally all determined, thereby its emotion composition is more stable, can be used as a snatch of music.But long and divide chapters and sections, the music between each chapters and sections to have different emotions for the time, for example symphony etc. then can not be as a snatch of music.

S20: the emotion composition of analyzing snatch of music: determine the emotion disaggregated model, and analyze the emotion composition of described snatch of music according to described emotion disaggregated model, determine the shared number percent of each emotion composition in the described snatch of music, obtain with described number percent be element, the music emotion vector that is associated with described snatch of music.

The emotion disaggregated model has multiple, according to different standards, multiple different disaggregated model can be arranged.For example common have Hevner emotion ring disaggregated model, a Thayer thymopsyche model.Hevner emotion ring disaggregated model is divided into sacredness, sadness, yearning, lyric, slim and graceful, happy, enthusiasm and 8 kinds of life (as shown in Figure 2) with the emotion of music.Thayer thymopsyche model is divided into the emotion of music vigor-and happy, vigor-sentiment, tranquil-happy and tranquil-sentimental 4 kinds arranged.

After determining the emotion disaggregated model, can analyze the emotion composition of a snatch of music, can obtain based on certain emotion constituent analysis algorithm according to the emotion disaggregated model.For a definite emotion disaggregated model, constitute a music emotion vector as component with all affective styles in this disaggregated model.The analysis that a snatch of music is carried out the emotion composition promptly is the shared number percent of component of determining each affective style, and is that element obtains the music emotion vector that is associated with snatch of music with described number percent.

Based on different emotion disaggregated models, obtaining the music emotion vector has diverse ways.Present embodiment is an example with Hevner emotion ring, describes the process of the music emotion vector that obtains a snatch of music.

As shown in Figure 2, be Hevner emotion ring synoptic diagram.Hevner describes the different emotions attribute of 67 descriptive adjective worlds of art of vocabulary utilization, is included into the type that 8 near synonym bunch expressions belong to close emotional responses according to classification, has constituted an annular according to its mutual relation, is called Hevner emotion ring.Wherein any one link all with its before and after adjacent link on affective logic, have progressive relationship, promptly think adjacent emotion before or after in the emotional change of the rational faculty, can being smoothly transitted into it.

By to the emotional semantic Study on similarity, obtain following emotion similar matrix about music emotion Semantic Similarity tolerance:

Music emotion is natural has a composite attribute, this be the caused people of diversity by the music sensation to the mixed feeling cognitive reaction that music took place, isolated with language value or the linguistic expression this composite attribute that all is beyond expression.It is different that different people is held for the emotion intension of music, but this difference is deferred to the constraint of Hevner emotion ring, and the similarity matrix of emotional semantic can be represented this inherent constraint.

Obtain 8 emotion base unit weight e according to above-mentioned matrix with its 8 row vectors _i(i=1～8).

Be defined as follows computing for emotion music vector:

Additive operation: if A, B is 2 music emotion vectors, a _i, b _i(i=1 2...8) is element in the vector respectively, β ₁, β ₂Be positive constant, then the additive operation of music emotion vector is defined as

${\mathrm{\β}}_{1}A+{\mathrm{\β}}_{2}B=\frac{1}{{\mathrm{\β}}_1+{\mathrm{\β}}_2}{({\mathrm{\β}}_1{a}_i+{\mathrm{\β}}_2{b}_i)}_{1\×8}$

Logical operation:

$\⫬A={(1-a_i)}_{1\×8}$

A∨B＝(a _i∨b _i) _1×8

A∧B＝(a _i∧b _i) _1×8

And following De Morgan rule is satisfied in logical operation:

1∧A＝A

0∨A＝A

$\⫬\⫬A=A$

The additive operation of music emotion vector is to logical operation ∧, and ∨ satisfies following law of distribution:

A+(B∨C)＝(A+B)∨(A+C)

A+(B∧C)＝(A+B)∧(A+C)

And further define subtraction by above-mentioned definition:

${\mathrm{\β}}_{1}A-{\mathrm{\β}}_{2}B={\mathrm{\β}}_{1}A+{\mathrm{\β}}_2(\⫬B)$

If w=[w so ₁, w ₂... w ₈] be the Fuzzy Distribution of certain song about 8 emotion components in the Hevner emotion ring, then can obtain the formula of following computational music emotion vector, wherein e according to the additive operation of emotion vector _iBe the emotion base vector.

$E_K=\underset{i=1}{\overset{8}{\mathrm{\Σ}}}{w}_i{e}_i$

S30: the main emotion composition of determining snatch of music: with the emotion composition that accounts for the largest percentage main emotion composition as described snatch of music.Hevner emotion ring has emotion composition in 8, and therefore different music can be taken the emotion composition as the leading factor with these 8 kinds of emotion compositions.

S40: judge whether handled music is last song, is then to be for further processing, otherwise forward step S10 to.

S50: all music are classified: the identical snatch of music of main emotion composition is divided into a class.Therefore all music can be divided into 8 classes as if being as the criterion, all music can be divided into 4 classes as if being as the criterion with Thayer thymopsyche model with Hevner emotion ring model.

S60: the emotion distance between the computational music: calculate distance between each snatch of music to belonging to of a sort snatch of music.Below with the music emotion vector representation be:

P＝[a ₁，a ₂，...，a _s] (1)

Wherein s is an emotion classification number, a _iRepresent i class emotion composition, i=1 ..., s.According to the emotion composition of music, judge its main emotion constituent class I (maximum composition):

$I=\arg \left(\underset{i}{\mathrm{Max}}\left(a_i\right)\right)---\left(2\right)$

I=1 wherein ..., s.

To I class emotion,, constitute the two-dimensional matrix D of M * M dimension to all music calculating distance each other wherein _I, wherein M is the number of fragment in such.D (i, j) _IRepresent i snatch of music X in the I class emotion _iWith j snatch of music X _jDistance:

D(i，j) _I＝dis(P _i，P _j) (3)

X wherein _i, X _jThe music of ∈ I class emotion, dis is the function of computed range, P _i, P _jBe respectively X _i, X _jThe emotion composition:

P _i＝[a _i，1，a _i，2，...，a _i，s，] (4)

P _j＝[a _j，1，a _j，2，...，a _j，s，] (5)

The employing Euclidean distance is represented the emotion distance between the music in the present embodiment, then:

$\mathrm{dis}(P_i,P_j)=\sqrt{\underset{m=1}{\overset{s}{\mathrm{\Σ}}}{(a_{i,m}-b_{j,m})}^2}---\left(6\right)$

Therefore, matrix D _IFollowing character is arranged:

D(i，j) _I＝D(j，i) _I (7)

D(i，i) _I＝0 (8)

Wherein, i=1,2 ..., M; J=1,2 ..., M.

Distance table D _IAnd music has wherein promptly constituted I class database.

To arbitrary song, determine after its main emotion composition classification I, according to distance matrix D _ICan obtain itself and all distances of similar other music.Can find and the very high music of this song emotion similarity apart from after sorting these.

The above embodiment has only expressed several embodiment of the present invention, and it describes comparatively concrete and detailed, but can not therefore be interpreted as the restriction to claim of the present invention.Should be pointed out that for the person of ordinary skill of the art without departing from the inventive concept of the premise, can also make some distortion and improvement, these all belong to protection scope of the present invention.Therefore, the protection domain of patent of the present invention should be as the criterion with claims.

Claims (3)

1. the balancing method of a music emotion similarity is characterized in that, comprises the steps:

Receive snatch of music;

Analyze the emotion composition of snatch of music: determine the emotion disaggregated model, and analyze the emotion composition of described snatch of music according to described emotion disaggregated model, determine the shared number percent of each emotion composition in the described snatch of music, obtain with described number percent be element, the music emotion vector that is associated with described snatch of music;

Determine the main emotion composition of snatch of music: with the emotion composition that accounts for the largest percentage main emotion composition as described snatch of music;

Repeat above-mentioned steps all music are carried out same processing;

Snatch of music is classified: the identical snatch of music of main emotion composition is divided into a class;

Emotion distance between the computational music fragment: to belonging to the distance between the music emotion vector that of a sort snatch of music calculates with each snatch of music is associated;

Determine the music emotion similarity according to the emotion distance between snatch of music: emotion distance is big more, and similarity is more little; Emotion distance is more little, and similarity is big more.

2. the balancing method of music emotion similarity as claimed in claim 1 is characterized in that, described emotion disaggregated model is Hevner emotion ring model or thayer thymopsyche model.

3. the balancing method of music emotion similarity as claimed in claim 1 is characterized in that, the emotion distance between the computational music fragment comprises the Euclidean distance that calculates between the music emotion vector be associated with snatch of music, promptly use D (i, j) _IRepresent i snatch of music X in the I class emotion _iWith j snatch of music X _jDistance:

D(i，j) _I＝dis(P _i，P _j)

X wherein _i, X _jThe music of ∈ I class emotion, dis is the function of computed range, P _i, P _jBe respectively X _i, X _jThe emotion composition, s is the emotion classification number in the emotion disaggregated model that adopts:

P _i＝[a _i，1，a _i，2，...，a _i，s，]

P _j＝[a _j，1，a _j，2，...，a _j，s，]

Then the Euclidean distance between the music emotion vector calculates with following formula:

$\mathrm{dis}(P_i,P_j)=\sqrt{\underset{m=1}{\overset{s}{\mathrm{\Σ}}}{(a_{i,m}-b_{j,m})}^2}.$

Images